Prior to this commit, the `MediaType` and `MimeType` "copy" constructors
would not leverage the fact that the existing instance has been
validated already (types, subtype and parameters have been checked
already for errors) and the entire validation would be performed again.
This would also allocate map instances in the process.
This commit ensures that the already validated information is reused
directly and that we avoid unnessecary operations and allocations for
such constructors.
Closes gh-36318
This is not actually triggered on 6.2.x but nevertheless worth aligning.
Includes fix for return type declaration in PropertyAccessor subclasses.
Includes related polishing from main commits.
See gh-36024
Add warnings to the class-level Javadoc for MergedAnnotations,
AnnotatedTypeMetadata, AnnotationMetadata, and MethodMetadata to point
out that annotations may be ignored if their attributes reference types
that are not present in the classpath.
Closes gh-35959
(cherry picked from commit b916dc962e)
Read the respective fields only once in the values(), entrySet(), and
keySet() methods.
Closes gh-35822
Signed-off-by: Patrick Strawderman <pstrawderman@netflix.com>
(cherry picked from commit 3b6be3d4d3)
The Spliterators returned by values, entrySet, and keySet incorrectly
reported the SIZED characteristic, instead of CONCURRENT. This could
lead to bugs when the map is concurrently modified during a stream
operation.
For keySet and values, the incorrect characteristics are inherited from
AbstractMap, so to rectify that the respective methods are overridden,
and custom collections are provided that report the correct Spliterator
characteristics.
Closes gh-35817
Signed-off-by: Patrick Strawderman <pstrawderman@netflix.com>
(cherry picked from commit ed75906834)
This commit ensures that both `VirtualThreadDelegate` implementations
expose the same public API. If not, JAR verification fails with the
following message:
```
jar --validate --file spring-core-6.2.13-SNAPSHOT.jar
entry: META-INF/versions/21/org/springframework/core/task/VirtualThreadDelegate.class, contains a class with different api from earlier version
```
Fixes gh-35773
When JAR manifest Class-Path entries contain absolute paths (as Gradle
creates on Windows for long classpaths), PathMatchingResourcePatternResolver
incorrectly rejected them.
Fixes#35730
Signed-off-by: Artur Signell <artur@vaadin.com>
When concurrency limiting is enabled via setConcurrencyLimit() and
thread creation fails in doExecute() (e.g., OutOfMemoryError from
Thread.start()), the concurrency permit acquired by beforeAccess()
is never released because TaskTrackingRunnable.run() never executes.
This causes the concurrency count to permanently remain at the limit,
causing all subsequent task submissions to block forever in
ConcurrencyThrottleSupport.onLimitReached().
Root cause:
- beforeAccess() increments concurrencyCount
- doExecute() throws Error before thread starts
- TaskTrackingRunnable.run() never executes
- afterAccess() in finally block never called
- Concurrency permit permanently leaked
Solution:
Wrap doExecute() in try-catch block in the concurrency throttle path
and call afterAccess() in catch block to ensure permit is always
released, even when thread creation fails.
The fix only applies to the concurrency throttle path. The
activeThreads-only path does not need fixing because it never calls
beforeAccess(), so there is no permit to leak.
Test approach:
The test simulates thread creation failure and verifies that a
subsequent execution does not deadlock. The first execution should
fail with some exception (type doesn't matter), and the second
execution should complete within timeout if the permit was properly
released.
Signed-off-by: Park Juhyeong <wngud5957@naver.com>
Prior to this commit, getPubliclyAccessibleMethodIfPossible() in
ClassUtils incorrectly returned a hidden static method as an
"equivalent" method for a static method with the same signature;
however, a static method cannot be overridden and therefore has no
"equivalent" method in a super type.
To fix that bug, this commit immediately aborts the search for an
"equivalent" publicly accessible method when the original method is a
static method.
See gh-33216
See gh-35189
See gh-35556
Closes gh-35667
As reported in gh-34651, `DataBuffer#getByte` can be inefficient for
some implementations, as bound checks are performed for each call.
This commit introduces a new `forEachByte` method that helps with
traversing operations without paying the bound check cost for each byte.
Closes gh-35623
Prior to this commit, annotations were not found on parameters in an
overridden method unless the method was public. Specifically, the
search algorithm in AnnotatedMethod did not consider a protected or
package-private method in a superclass to be a potential override
candidate. This affects parameter annotation searches in
spring-messaging, spring-webmvc, spring-webflux, and any other
components that use or extend AnnotatedMethod.
To address that, this commit revises the search algorithm in
AnnotatedMethod to consider all non-final declared methods as potential
override candidates, thereby aligning with the search logic in
AnnotationsScanner for the MergedAnnotations API.
Closes gh-35349
Prior to this commit, gh-32097 added native support for Jetty for both
client and server integrations. The `JettyDataBufferFactory` was
promoted as a first class citizen, extracted from a private class in the
client support. To accomodate with server-side requirements, an extra
`buffer.retain()` call was performed.
While this is useful for server-side support, this introduced a bug in
the data buffer factory, as wrapping an existing chunk means that this
chunk is already retained.
This commit fixes the buffer factory implementation and moved existing
tests from mocks to actual pooled buffer implementations from Jetty.
The extra `buffer.retain()` is now done from the server support, right
before wrapping the buffer.
Fixes gh-35319